mindspore/ccsrc/backend/kernel_compiler/tbe/tbe_adapter.cc

@@ -87,6 +87,7 @@ static std::map<string, string> tbe_func_adapter_map = {
   {"apply_centered_rms_prop", "apply_centered_rms_prop_d"},
   {"transpose", "transpose_d"},
   {"fill", "fill_d"},
+  {"i_fmr", "ifmr"},
   {"unsorted_segment_sum", "unsorted_segment_sum_d"},
   {"unsorted_segment_prod", "unsorted_segment_prod_d"},
   {"concat", "concat_d"},

mindspore/ops/_op_impl/tbe/__init__.py

@@ -281,6 +281,7 @@ from .broadcast_to import _broadcast_to_tbe
 from .strided_read import _strided_read_tbe
 from .strided_write import _strided_write_tbe
 from .range import _range_tbe
+from .ifmr import _ifmr_tbe
 from .fused_mul_add_n_l2loss import _fused_mul_add_n_l2loss_tbe
 from .fused_mul_apply_momentum_extern import _fused_mul_apply_momentum_extern_tbe
 from .lamb_next_right import _lamb_next_right_tbe

mindspore/ops/_op_impl/tbe/ifmr.py

@@ -0,0 +1,47 @@
+# Copyright 2020 Huawei Technologies Co., Ltd
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ============================================================================
+
+"""IFMR op"""
+from mindspore.ops.op_info_register import op_info_register, TBERegOp, DataType
+
+ifmr_op_info = TBERegOp("IFMR") \
+    .fusion_type("OPAQUE") \
+    .async_flag(False) \
+    .binfile_name("ifmr.so") \
+    .compute_cost(10) \
+    .kernel_name("ifmr") \
+    .partial_flag(True) \
+    .attr("min_percentile", "required", "float", "all") \
+    .attr("max_percentile", "required", "float", "all") \
+    .attr("search_range", "required", "listFloat", "all") \
+    .attr("search_step", "required", "float", "all") \
+    .attr("with_offset", "required", "bool", "all") \
+    .input(0, "data", False, "required", "all") \
+    .input(1, "data_min", False, "required", "all") \
+    .input(2, "data_max", False, "required", "all") \
+    .input(3, "cumsum", False, "required", "all") \
+    .output(0, "scale", False, "required", "all") \
+    .output(1, "offset", False, "required", "all") \
+    .dtype_format(DataType.F16_Default, DataType.F16_Default, DataType.F16_Default, DataType.I32_Default,
+                  DataType.F16_Default, DataType.F16_Default) \
+    .dtype_format(DataType.F32_Default, DataType.F32_Default, DataType.F32_Default, DataType.I32_Default,
+                  DataType.F32_Default, DataType.F32_Default) \
+    .get_op_info()
+
+
+@op_info_register(ifmr_op_info)
+def _ifmr_tbe():
+    """IFMR TBE register"""
+    return

mindspore/ops/operations/__init__.py

@@ -53,7 +53,7 @@ from .math_ops import (Abs, ACos, Asin, Asinh, AddN, AccumulateNV2, AssignAdd, A
                        NPUAllocFloatStatus, NPUClearFloatStatus,
                        NPUGetFloatStatus, Pow, RealDiv, IsNan, IsInf, IsFinite, FloatStatus,
                        Reciprocal, CumSum, HistogramFixedWidth, SquaredDifference, Xdivy, Xlogy,
-                       Sin, Sqrt, Rsqrt, BesselI0e, BesselI1e, TruncateDiv, TruncateMod,
+                       Sin, Sqrt, Rsqrt, BesselI0e, BesselI1e, TruncateDiv, TruncateMod, IFMR,
                        Square, Sub, TensorAdd, Sign, Round, SquareSumAll, Atan, Atanh, Cosh, Sinh, Eps, Tan)
 
 from .random_ops import (RandomChoiceWithMask, StandardNormal, Gamma, Poisson, UniformInt, UniformReal,
@@ -104,6 +104,7 @@ __all__ = [
     'AddN',
     'AccumulateNV2',
     'Sub',
+    'IFMR',
     'CumSum',
     'MatMul',
     'BatchMatMul',

mindspore/ops/operations/math_ops.py

@@ -3514,3 +3514,64 @@ class Eps(PrimitiveWithInfer):
             'dtype': input_x['dtype'],
         }
         return out
+
+
+class IFMR(PrimitiveWithInfer):
+    """
+    The TFMR(Input Feature Map Reconstruction).
+
+    Args:
+        min_percentile (float): Min init percentile.
+        max_percentile (float): Max init percentile.
+        search_range Union[list(float), tuple(float)]: Range of searching.
+        search_step (float): Step size of searching.
+        with_offset (bool): Whether using offset.
+
+    Inputs:
+        - **data** (Tensor) - A Tensor of feature map. With float16 or float32 data type.
+        - **data_min** (Tensor) - A Tensor of min value of feature map, the shape is :math:`(1)`.
+          With float16 or float32 data type.
+        - **data_max** (Tensor) - A Tensor of max value of feature map, the shape is :math:`(1)`.
+          With float16 or float32 data type.
+        - **cumsum** (Tensor) - A `1-D` Tensor of cumsum bin of data. With int32 data type.
+
+    Outputs:
+        - **scale** (Tensor) - A tensor of optimal scale, the shape is :math:`(1)`. Data dtype is float32.
+        - **offset** (Tensor) - A tensor of optimal offset, the shape is :math:`(1)`. Data dtype is float32.
+
+    Examples:
+        >>> data = Tensor(np.random.rand(1, 3, 6, 4).astype(np.float32))
+        >>> data_min = Tensor([0.1], mstype.float32)
+        >>> data_max = Tensor([0.5], mstype.float32)
+        >>> cumsum = Tensor(np.random.rand(4).astype(np.int32))
+        >>> ifmr = P.IFMR(min_percentile=0.2, max_percentile=0.9, search_range=(1.0, 2.0),
+                          search_step=1.0, with_offset=False)
+        >>> output = ifmr(data, data_min, data_max, cumsum)
+    """
+
+    @prim_attr_register
+    def __init__(self, min_percentile, max_percentile, search_range, search_step, with_offset):
+        validator.check_value_type("min_percentile", min_percentile, [float], self.name)
+        validator.check_value_type("max_percentile", max_percentile, [float], self.name)
+        validator.check_value_type("search_range", search_range, [list, tuple], self.name)
+        for item in search_range:
+            validator.check_float_positive("item of search_range", item, self.name)
+        validator.check('search_range[1]', search_range[1], 'search_range[0]', search_range[0], Rel.GE, self.name)
+        validator.check_value_type("search_step", search_step, [float], self.name)
+        validator.check_value_type("offset_flag", with_offset, [bool], self.name)
+
+    def infer_shape(self, data_shape, data_min_shape, data_max_shape, cumsum_shape):
+        validator.check_integer("dims of data_min", len(data_min_shape), 1, Rel.EQ, self.name)
+        validator.check_integer("data_min[0]", data_min_shape[0], 1, Rel.EQ, self.name)
+        validator.check_integer("dims of data_max", len(data_max_shape), 1, Rel.EQ, self.name)
+        validator.check_integer("data_max[0]", data_max_shape[0], 1, Rel.EQ, self.name)
+        validator.check_integer("dims of cumsum", len(cumsum_shape), 1, Rel.EQ, self.name)
+        return (1,), (1,)
+
+    def infer_dtype(self, data_dtype, data_min_dtype, data_max_dtype, cumsum_dtype):
+        valid_types = [mstype.float32, mstype.float16]
+        validator.check_tensor_type_same({"input_value": data_dtype}, valid_types, self.name)
+        validator.check_tensor_type_same({"input_min": data_min_dtype}, valid_types, self.name)
+        validator.check_tensor_type_same({"input_max": data_max_dtype}, valid_types, self.name)
+        validator.check_tensor_type_same({"input_bins": cumsum_dtype}, [mstype.int32], self.name)
+        return mstype.tensor_type(mstype.float32), mstype.tensor_type(mstype.float32)

tests/ut/python/ops/test_ops.py

@@ -1275,6 +1275,13 @@ test_case_math_ops = [
         'block': P.Mod(),
         'desc_inputs': [[3, 4, 5], [2, 3, 4, 5]],
         'desc_bprop': [[2, 3, 4, 5]]}),
+    ('IFMR', {
+        'block': P.IFMR(min_percentile=0.2, max_percentile=0.9, search_range=(1.0, 2.0),
+                        search_step=1.0, with_offset=False),
+        'desc_inputs': [[3, 4, 5], Tensor([0.1], mstype.float32), Tensor([0.9], mstype.float32),
+                        Tensor(np.random.rand(4).astype(np.int32))],
+        'desc_bprop': [],
+        'skip': ['backward']}),
 ]
 
 test_case_nn_ops = [